As suggested by Elqayam & Evans (E&E), normativism faces an arbitration problem when many researchers in the field are familiar with many formalisms that are eligible for normative status. Normativism, obviously, faces even more of a problem when very few researchers in the field have heard of any eligible formalism. Accordingly, normativism is expected to be strong in fields where exactly one formalism is a well-known candidate for normative status. The psychology of reasoning qualifies for this category, and so does the psychology of decision-making. Although there are investigations of formalisms that provide alternative norms for reasoning (Benferhat et al. Reference Benferhat, Bonnefon, Da and Neves2005) or decision-making (Bonnefon et al. Reference Bonnefon, Dubois, Fargier and Leblois2008), they have garnered considerably less attention than the dominant competitors: logic and Bayesianism (for reasoning), and utility maximization (for decision-making).
An interesting trend is developing, however, with the ongoing convergence of the two fields. Students of reasoning, in particular, are increasingly incorporating decision-theoretic elements in their investigations. Whereas reasoning research has traditionally focused on the derivation of disinterested beliefs from other disinterested beliefs, current research is giving increasing attention to reasoning about preferences and decisions. Under its most general form, the problem is to account for the ways in which people infer the beliefs, preferences, or decisions of other people based on partial knowledge about this whole set of information. Because there is no single, well-entrenched formal model of how this should be done, the drive toward normativism is paradoxically weaker when investigating how people reason about decisions, than when investigating either how they reason in general, or how they make decisions. Even in Artificial Intelligence, models of how agents reason about decisions make a clear distinction between their own decision-making models, the models others may be using, and the extent to which everyone may deviate from these models (Gal & Pfeffer Reference Gal and Pfeffer2008).
Accordingly, when moving from reasoning about facts to reasoning about decisions, E&E's distinction between prescriptive and descriptive normativism unpacks into three distinct questions: (1) Do models of reasoning about decisions consider that people assume other people to be normative decision-makers? (2) Is there an optimal way to reason about decisions, whether or not other people are assumed to be normative decision-makers? (3) If so, does reasoning about decisions reflect this optimal solution? The answers to these three questions can vary to a large extent in different psychological approaches to reasoning about decisions. The computational model offered by Baker et al. (Reference Baker, Saxe and Tenenbaum2009), and the descriptive model put forward by Bonnefon (Reference Bonnefon2009) may represent two extremes on that spectrum, while the graphical causal model approach adopted by Hagmayer and Sloman (Reference Hagmayer and Sloman2009) seems to take an intermediate perspective.
On the normativist side of the spectrum, Baker et al. (Reference Baker, Saxe and Tenenbaum2009) offer a model of reasoning about decisions where the response to each of the three questions above is “Yes”. In this model, other people are assumed to be rational decision-makers who choose actions that achieve their desires most efficiently, given their beliefs. By virtue of being computational, the model assumes that there is an optimal solution to the problem of reasoning about decisions (prescriptive normativism), and the normative framework that it uses is Bayesian inverse planning from Markov decision problems. Finally, it is expected that real-life inferences will be better predicted by this optimal model than by simple heuristics (descriptive normativism).
Hagmayer and Sloman (Reference Hagmayer and Sloman2009, Experiment 4) investigated inferences about the beliefs of other people, based on their preferences and decisions. Their core assumption is that people assume others to conceive of choices as interventions on a causal graphical network. Hagmayer and Sloman do not explicitly state that the choice-as-intervention (CI) principle is normative, but the reasons they consider for someone not to endorse it have an obvious non-normative flavor (e.g., self-deception, self-handicapping, illusion of control, superstition). Thus, the model appears to feature a normative calculus (causal graphical models) involving the assumption that other people are normative decision-makers (the CI principle). As such, it would qualify for prescriptive normativism, as defined by E&E. Hagmayer and Sloman do not endorse descriptive normativism, however, as they explicitly consider the conditions (e.g., lack of motivation or capacity) under which inferences are not expected to reflect the CI principle and the causal graphical calculus.
Finally, at the descriptive end of the spectrum, Bonnefon (Reference Bonnefon2009) offers a theory of how reasoners infer the decisions and attitudes of agents, based on utility conditional statements such as “if p then q”, where p and q bear on the utility functions of these and other agents. Rather than assuming people to believe that other agents are fully rational, the theory is based on folk axioms of decision, that is, the naïve beliefs that people have about the way other agents make their decisions. The theory is purely descriptive in the sense that it does not assume the existence of an optimal solution for reasoning about utility conditionals. Its response to each of the three questions above is therefore “No”.
As can be seen from these three examples, the new field of reasoning about decisions is remarkably opened to descriptivist as well as normativist approaches, in contrast to its two parent fields (reasoning and decision-making), which are much more influenced by normativist approaches. In light of the case that E&E make against the dominance of normativist approaches, this flexibility bodes well for the future of the psychology of reasoning about decisions.
As suggested by Elqayam & Evans (E&E), normativism faces an arbitration problem when many researchers in the field are familiar with many formalisms that are eligible for normative status. Normativism, obviously, faces even more of a problem when very few researchers in the field have heard of any eligible formalism. Accordingly, normativism is expected to be strong in fields where exactly one formalism is a well-known candidate for normative status. The psychology of reasoning qualifies for this category, and so does the psychology of decision-making. Although there are investigations of formalisms that provide alternative norms for reasoning (Benferhat et al. Reference Benferhat, Bonnefon, Da and Neves2005) or decision-making (Bonnefon et al. Reference Bonnefon, Dubois, Fargier and Leblois2008), they have garnered considerably less attention than the dominant competitors: logic and Bayesianism (for reasoning), and utility maximization (for decision-making).
An interesting trend is developing, however, with the ongoing convergence of the two fields. Students of reasoning, in particular, are increasingly incorporating decision-theoretic elements in their investigations. Whereas reasoning research has traditionally focused on the derivation of disinterested beliefs from other disinterested beliefs, current research is giving increasing attention to reasoning about preferences and decisions. Under its most general form, the problem is to account for the ways in which people infer the beliefs, preferences, or decisions of other people based on partial knowledge about this whole set of information. Because there is no single, well-entrenched formal model of how this should be done, the drive toward normativism is paradoxically weaker when investigating how people reason about decisions, than when investigating either how they reason in general, or how they make decisions. Even in Artificial Intelligence, models of how agents reason about decisions make a clear distinction between their own decision-making models, the models others may be using, and the extent to which everyone may deviate from these models (Gal & Pfeffer Reference Gal and Pfeffer2008).
Accordingly, when moving from reasoning about facts to reasoning about decisions, E&E's distinction between prescriptive and descriptive normativism unpacks into three distinct questions: (1) Do models of reasoning about decisions consider that people assume other people to be normative decision-makers? (2) Is there an optimal way to reason about decisions, whether or not other people are assumed to be normative decision-makers? (3) If so, does reasoning about decisions reflect this optimal solution? The answers to these three questions can vary to a large extent in different psychological approaches to reasoning about decisions. The computational model offered by Baker et al. (Reference Baker, Saxe and Tenenbaum2009), and the descriptive model put forward by Bonnefon (Reference Bonnefon2009) may represent two extremes on that spectrum, while the graphical causal model approach adopted by Hagmayer and Sloman (Reference Hagmayer and Sloman2009) seems to take an intermediate perspective.
On the normativist side of the spectrum, Baker et al. (Reference Baker, Saxe and Tenenbaum2009) offer a model of reasoning about decisions where the response to each of the three questions above is “Yes”. In this model, other people are assumed to be rational decision-makers who choose actions that achieve their desires most efficiently, given their beliefs. By virtue of being computational, the model assumes that there is an optimal solution to the problem of reasoning about decisions (prescriptive normativism), and the normative framework that it uses is Bayesian inverse planning from Markov decision problems. Finally, it is expected that real-life inferences will be better predicted by this optimal model than by simple heuristics (descriptive normativism).
Hagmayer and Sloman (Reference Hagmayer and Sloman2009, Experiment 4) investigated inferences about the beliefs of other people, based on their preferences and decisions. Their core assumption is that people assume others to conceive of choices as interventions on a causal graphical network. Hagmayer and Sloman do not explicitly state that the choice-as-intervention (CI) principle is normative, but the reasons they consider for someone not to endorse it have an obvious non-normative flavor (e.g., self-deception, self-handicapping, illusion of control, superstition). Thus, the model appears to feature a normative calculus (causal graphical models) involving the assumption that other people are normative decision-makers (the CI principle). As such, it would qualify for prescriptive normativism, as defined by E&E. Hagmayer and Sloman do not endorse descriptive normativism, however, as they explicitly consider the conditions (e.g., lack of motivation or capacity) under which inferences are not expected to reflect the CI principle and the causal graphical calculus.
Finally, at the descriptive end of the spectrum, Bonnefon (Reference Bonnefon2009) offers a theory of how reasoners infer the decisions and attitudes of agents, based on utility conditional statements such as “if p then q”, where p and q bear on the utility functions of these and other agents. Rather than assuming people to believe that other agents are fully rational, the theory is based on folk axioms of decision, that is, the naïve beliefs that people have about the way other agents make their decisions. The theory is purely descriptive in the sense that it does not assume the existence of an optimal solution for reasoning about utility conditionals. Its response to each of the three questions above is therefore “No”.
As can be seen from these three examples, the new field of reasoning about decisions is remarkably opened to descriptivist as well as normativist approaches, in contrast to its two parent fields (reasoning and decision-making), which are much more influenced by normativist approaches. In light of the case that E&E make against the dominance of normativist approaches, this flexibility bodes well for the future of the psychology of reasoning about decisions.